Hanging rail inspection robot integrated with hanging rail cleaning and pig house ventilation dynamic matching system and method for pig herd thermal imaging inspection

Abstract

The application discloses a pig house ventilation dynamic matching system and method for pig group thermal imaging inspection, and belongs to the field of livestock and poultry breeding intelligent equipment and Internet of Things automatic control technology. The overhead rail cleaning mechanism adopts an elastic floating structure, can clean the overhead rail running surface synchronously with the inspection robot, and avoids dirt accumulation and line walking failure. The inspection robot is equipped with a thermal imaging module, an environment sensing module and a visible light sensing module. The system performs multi-scale non-uniform correction, CBAM attention body surface temperature extraction, extended Kalman multi-frame temperature fusion, calculates the sensible temperature in combination with a pig heat balance equation, constructs a grid thermal load distribution map, multi-dimensionally judges a high heat load area, and can dispatch the robot to carry out intensive sampling. Fuzzy PID or MPC control is adopted to realize partitioned air volume distribution, a permanent magnet synchronous fan is driven through feedforward-feedback compound control, and periodic rolling iteration is adopted to realize pig house ventilation dynamic closed-loop regulation and control, so that the cold and heat stress of the pig group is reduced, and energy consumption is reduced.

Description

Technical Field

[0001] This invention belongs to the field of intelligent equipment and IoT-based automated control technology for livestock and poultry farming, specifically relating to a hanging rail inspection robot that integrates hanging rail cleaning and a dynamic matching system and method for pig house ventilation based on thermal imaging inspection of pig herds. Background Technology

[0002] Large-scale, enclosed pigsties are the mainstream model for modern pig farming. The internal environment of the pigsty directly determines the health level, stress level, and production performance of the pigs. Enclosed pigsties generally use overhead rail-mounted inspection robots. These robots, equipped with various sensors, move back and forth along the rails to achieve comprehensive environmental inspection and pig status monitoring throughout the pigsty.

[0003] However, existing technologies still have several problems in practical applications: First, the overhead rails are prone to accumulating dirt, resulting in poor reliability of the inspection robot. Feed scraps, pig hair, dust, and flocculent suspended matter are continuously generated inside pigsties, settling and accumulating on the overhead rails over time. Existing overhead rail inspection robots lack a track cleaning mechanism; the accumulation of debris on the track surface causes wheel slippage, increased walking resistance, and positioning deviation. In severe cases, wheel jamming and robot derailment may occur. Manually cleaning the overhead rails at height is difficult and risky, making regular real-time cleaning impossible; only periodic shutdowns for manual cleaning are possible, reducing the effective operating time of the inspection equipment.

[0004] Secondly, traditional pigsty ventilation systems suffer from inaccurate sensing methods: Traditional pigsty ventilation mostly relies on fixed-point temperature and humidity sensors to collect air environment parameters, and on asynchronous motor fans for simple threshold start-stop control. Fixed-point sensors can only collect air temperature at local points and cannot reflect the true body temperature felt by pigs; the body temperature felt by pigs is affected by multiple factors coupled together, including surface temperature, air temperature and humidity, ambient wind speed, pig density, and pig behavior and posture. Relying solely on air temperature can easily lead to misjudgments of heat or cold stress, causing heat stress or cold stress in pigs.

[0005] Third, the application of thermal imaging temperature measurement in the complex environment of pigsties presents multiple technical challenges. Traditional methods for collecting pig surface temperature using thermal imaging require manual selection of the Region of Interest (ROI). Pigs exhibit various postures, including lying down, side-lying, and standing, and there are also localized obstructions, making it impossible for manual ROI selection to automatically adapt to these posture changes. Furthermore, the heating lamps, water pipes, and equipment heat dissipation within the pigsty generate significant heat sources, leading to ghosting noise and fixed pattern noise in the thermal imaging images, causing measurement inaccuracies. Single-frame thermal images are also affected by airflow disturbances and localized obstructions, resulting in large fluctuations and unstable measurement results. Existing technologies lack differentiated weighting for key physiological temperature measurement sites such as the base of the ears, the midline of the back, and the groin, and do not incorporate the pig's thermal equilibrium physical model to invert the actual perceived temperature, making it difficult to directly use the measurement results for ventilation decisions.

[0006] Fourth, the ventilation control strategy is rigid and lacks spatial distribution dimension. Most existing ventilation controls use a uniform switch for the entire pigsty or simple PID control, without considering the differences in pig density, behavior, and local heat load distribution in different pen areas within the pigsty, and cannot generate a spatial heat load distribution map of the pigsty; it lacks multi-dimensional high heat load judgment logic and a zoned air volume allocation strategy without behavioral constraints, which easily leads to insufficient or excessive ventilation in some areas. Excessive ventilation can also create cross drafts, causing cold stress in pigs and harming animal welfare.

[0007] Fifth, the response of the fan actuator is lagging. Traditional asynchronous motor fans have a narrow speed range and slow response speed; the conversion from air volume to speed lacks feedforward compensation and relies solely on simple feedback PID. In response to transient heat load changes in pigsties, the tracking is lagging, and the air volume cannot match the dynamic heat load of the pigs in real time, resulting in energy waste and poor environmental control. Furthermore, most traditional air supply windows are single downward air supply, with the airflow concentrated and blowing directly onto the pens below, which can easily cause excessively high local wind speeds, insufficient fresh air supply to surrounding pens, and uneven airflow distribution in the pigsty, which can easily induce cold stress in pigs. Summary of the Invention

[0008] The purpose of this invention is to provide a suspended rail cleaning mechanism that solves the hardware problems of dust, pig hair, and feed debris accumulating on the running surface of pigsty suspended rails, making manual cleaning difficult, and causing the inspection robot's wheels to slip, jam, and derail. It also provides a suspended rail inspection robot that simultaneously performs inspection work and rail cleaning without requiring separate shutdown for cleaning, thus solving problems such as distortion from fixed-point environmental sensors, poor accuracy of thermal imaging temperature measurement in complex pigsty scenarios, and inaccurate calculation of perceived temperature using traditional empirical models. Furthermore, it provides a dynamic matching system and method for pigsty ventilation based on pig herd thermal imaging inspection. By constructing a grid-based heat load distribution map of the pigsty and integrating pig herd heat stress index, pig density, and behavioral activity, it achieves spatial-dimensional heat load characterization, solving the problems of rigid ventilation control strategies and delayed fan response.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A suspended rail cleaning mechanism includes a suspended rail fixed to the bottom surface of a pigsty ceiling. A horizontally positioned connecting seat is located at the bottom of the rail. Two sets of vertically positioned connecting seats, symmetrically distributed on the left and right sides of the rail, are fixed to the top surface of the connecting seat. Cleaning wheels, elastically contacting the rail's running surface, are rotatably mounted on the connecting seats. The cleaning mechanism utilizes the connecting seats to install the cleaning wheels, which are symmetrically arranged on the left and right sides of the rail and elastically abut against the rail's running surface. During the inspection robot's movement, the cleaning wheels continuously sweep the rail's running surface, scraping away accumulated pig hair, dust, and feed debris. This eliminates the need for manual high-altitude rail cleaning; the cleaning operation is completed simultaneously with the inspection, preventing rail contamination that could cause slippage or jamming of the inspection wheels and improving the robot's operational stability.

[0010] To achieve elastic contact of the sweeping wheel, a vertical groove is formed through the connecting vertical seat. A vertical guide rod is fixedly connected within the vertical groove, and a slide block is slidably mounted on the vertical guide rod. Limiting blocks are fixed on both sides of the slide block to restrict its rotation and allow it to slide only up and down. A top spring is fitted on the vertical guide rod to push the slide block downwards. The sweeping wheel is rotatably mounted on the inner side of the slide block. The vertical guide rod, slide block, limiting blocks, and top spring form an elastic floating structure. The limiting blocks restrict the slide block to slide only up and down along the vertical guide rod and prevent circumferential rotation. The top spring continuously pushes the slide block downwards, ensuring the sweeping wheel remains firmly pressed against the running surface of the overhead rail. When there are machining errors, installation deflection deformation, or large protruding debris on the rail surface, the sweeping wheel can float upwards to avoid these issues, preventing jamming or damage. This ensures continuous and effective contact between the sweeping wheel and the running surface, adapts to track deformation, and improves cleaning reliability.

[0011] A suspended rail inspection robot includes a main body that travels along a suspended rail. A front anti-collision energy-absorbing box and a rear anti-collision energy-absorbing box are sequentially installed at the front and rear ends of the main body. The aforementioned suspended rail cleaning mechanism is sequentially installed in front of the front anti-collision energy-absorbing box and behind the rear anti-collision energy-absorbing box. A sensing module and a communication module are installed at the bottom of the main body, and an edge computing and central control module is installed at the top. The suspended rail cleaning mechanism is located at the front and rear ends of the robot. When the robot moves forward, the front cleaning mechanism cleans the rail; when it moves backward, the rear cleaning mechanism cleans the rail. The entire reciprocating inspection process allows for simultaneous cleaning of the suspended rail surface. The front and rear anti-collision energy-absorbing boxes buffer the impact when the robot encounters obstacles, protecting the cleaning mechanism and the robot body. The sensing module at the bottom collects data on the pigs and the environment inside the pigsty, while the communication module handles data interaction. The edge computing and central control module at the top performs local data processing. The hardware layout is rational, integrating rail cleaning, motion sensing, and edge computing.

[0012] To achieve propulsion, the inspection robot's main body includes a main shell. Two sets of symmetrical drive wheels, moving along the suspended rail, are rotatably mounted on the front top of the main shell via a front support arm. Two sets of symmetrical driven wheels, moving along the suspended rail, are rotatably mounted on the rear top of the main shell via a rear support arm. A servo motor and a drive sprocket driven by the servo motor are installed inside the main shell. A driven sprocket, coaxially connected to the drive wheels, is concealed within the front support arm. The drive and driven sprockets are connected via chain drive. A battery, supplying power to all electrical components, is also installed within the main shell. This dual-wheel system, employing drive and driven wheels, is symmetrically arranged to ensure balanced force distribution during robot suspension and movement. Power is transmitted via a servo motor-drive sprocket-chain-driven sprocket configuration. The chain drive is located inside the support arm to prevent corrosion of transmission components by pigsty dust. The built-in battery powers all electrical components, supporting reciprocating suspended rail inspections without external cables. The robot features a compact structure, reliable transmission, and is suitable for harsh working conditions in high-dust pigsties.

[0013] To achieve perception, communication, and control, the perception module includes an infrared thermal imager, an environmental sensor group, and a visible light camera. The environmental sensor group includes at least a temperature and humidity sensor, an ammonia sensor, and a CO2 sensor. The perception module enables thermal imaging temperature field acquisition, environmental gas parameter acquisition, and pig herd image acquisition. The communication module includes an industrial switch and a remote monitoring terminal, which enables local networking and remote monitoring. The edge computing and central control module includes an edge computing unit, a central controller, and a data storage unit. The edge computing unit is responsible for image preprocessing and algorithm calculation, the central controller completes the ventilation decision logic, and the data storage unit stores thermal imaging frames, environmental time-series data, and control logs. Multi-source perception, edge algorithm calculation, and control logic are integrated into the inspection robot body, providing a complete perception and computing hardware foundation for subsequent dynamic matching of ventilation.

[0014] To facilitate heat dissipation, multiple heat dissipation windows are installed at the bottom of the main casing. These windows are divided into motor heat dissipation windows near the servo motors and battery heat dissipation windows near the batteries. The high temperature inside the pigsty causes the servo motors and batteries to continuously generate heat. The separate motor and battery heat dissipation windows provide targeted cooling for the power and energy storage components, reducing operating temperature rise, preventing motor overheating and power reduction, and avoiding battery degradation due to high temperatures. This improves the reliability of the robot operating continuously in the high-temperature environment of the pigsty.

[0015] A dynamic matching system for piggery ventilation based on thermal imaging inspection includes an air supply duct fixed to the ceiling of the piggery. A permanent magnet synchronous fan is installed at the entrance of the air supply duct to supply air into it. Multiple air supply windows that penetrate the ceiling and supply air to the pigs are installed on the bottom surface of the air supply duct. A hanging rail is fixed to the ceiling directly below the air supply duct. The aforementioned hanging rail inspection robot is installed on the hanging rail. A permanent magnet direct drive fan driver that can receive commands from the hanging rail inspection robot is connected to one side of the permanent magnet synchronous fan. Air outlets that supply air to four different directions are opened on the air supply windows. The air supply ducts are located above the ceiling, delivering air to the pig pens through air supply windows. A suspended rail is positioned directly below the air supply ducts on the ceiling, allowing an inspection robot to move along the rails and perform omnidirectional sensing. The permanent magnet direct-drive fan driver receives edge computing data from the inspection robot and control commands from the central control module, driving the permanent magnet synchronous fan with stepless speed regulation. The hardware system forms a complete closed-loop system of "mobile sensing - local computing - fan execution," achieving system integration of sensing hardware and ventilation execution hardware. Furthermore, the air supply windows are equipped with four air outlets facing different directions, evenly distributing airflow to the surrounding pens and preventing drafts caused by unidirectional airflow. It can simultaneously supply fresh air to multiple pen areas around the air supply windows, improving the uniformity of airflow distribution within the pens. Combined with zoned airflow adjustment, it can reduce cold stress caused by direct airflow to specific areas in the pigsty. The hardware system forms a complete closed-loop system of "mobile sensing - local computing - fan execution," achieving system integration of sensing hardware and ventilation execution hardware.

[0016] A method for dynamically matching pig house ventilation based on thermal imaging inspection of pig herds is disclosed. This method employs the aforementioned dynamic matching system for pig house ventilation based on thermal imaging inspection of pig herds, and its operational steps are as follows: S1. Multi-source data acquisition: The overhead inspection robot moves back and forth along the overhead rail, and the infrared thermal imager collects the thermal imaging temperature field of the pig herd as well as the background temperature field of the ground, walls and equipment; the environmental sensor group collects the spatial distribution of temperature, humidity, ammonia and CO2 parameters; the visible light camera collects images of the pig herd, identifies the pig density and posture, and calculates the behavioral activity level through optical flow method. S2. Thermal Imaging Multi-Algorithm Preprocessing: The edge computing unit performs multi-scale non-uniformity correction on the original thermal image; the pig's body surface temperature is adaptively extracted based on the CBAM attention mechanism, and differentiated temperature measurement weights are assigned to the ear root, back midline, groin, nose mirror, and tail root; then, the extended Kalman filter is used to perform optimal estimation of the body surface temperature of multiple consecutive frames, and output a stable pig body surface temperature; at the same time, Kriging interpolation is used to generate an air temperature interpolation field. S3. Calculation of perceived temperature and heat stress index: A perceived temperature model is constructed based on the physical equation of heat balance of pigs. The corresponding convective heat transfer coefficient-wind speed fitting relationship is selected according to the standing or lying posture of pigs. The membership degree of cold, comfort and heat stress is calculated by using fuzzy membership function, and the heat stress index TSI is solved. S4. Construct a heat load distribution map: Divide the pigsty space into m×n grid units; count the pig density and behavioral activity of each grid, and calculate the grid comprehensive heat load index; generate a heat load distribution surface through bicubic B-spline interpolation; maintain a cyclic buffer to store a multi-period historical heat load matrix, and calculate the heat load change rate as a basis for feedforward early warning. S5. Ventilation Decision and Zonal Airflow Allocation: Performs multi-dimensional high heat load judgment, combining absolute threshold, relative percentile, heat load rise rate, superimposed time window duration anti-jitter constraints, and identifies high heat load in special scenarios; calls a fuzzy PID controller or alternative MPC model predictive controller to calculate the total target airflow for the entire shed; performs zonal airflow allocation: obtains the baseline airflow according to the heat load ratio, applies upper and lower limit constraints on airflow based on the pig herd behavior status, completes total rebalancing, and applies smoothing constraints on airflow between adjacent zones; when ammonia concentration exceeds the threshold, a basic ventilation lower limit is forcibly set; S6. Closed-loop ventilation execution: The target air volume of each zone is converted into a permanent magnet synchronous fan speed command through feedforward-feedback composite control and sent to the driver; the permanent magnet synchronous fan performs speed regulation and transmits back the speed, current and vibration status. S7. Rolling Iteration: Enter the next round of inspection cycle, repeat steps S1-S6, and continuously dynamically match the ventilation volume of the pigsty. This method relies on the aforementioned hardware system and integrates a complete algorithm chain. It acquires spatial distribution data of the pigsty through multi-source mobile acquisition; suppresses ghosting noise from strong heat sources such as pigsty heating lamps through multi-scale non-uniform correction; uses the CBAM attention mechanism to extract body surface temperature without artificial ROI, combined with differentiated weights for key physiological parts of the pigs; uses extended Kalman filtering to eliminate instantaneous disturbances in single-frame images; generates a continuous air temperature field through Kriging interpolation; inverts the perceived temperature based on a thermal equilibrium physical model, distinguishes the convective heat transfer coefficient by posture, and obtains the heat stress index through fuzzy membership; constructs a grid-based heat load distribution map, and uses a historical matrix to calculate the rate of heat load change for feedforward early warning; uses multi-dimensional high heat load judgment with time-based anti-jittering to avoid false triggering; supports both fuzzy PID and MPC control modes; introduces behavioral constraints, total rebalancing, and adjacent smoothing constraints to suppress cross ventilation in zoned airflow allocation; forces basic ventilation when ammonia levels exceed limits; achieves rapid fan response through feedforward-feedback composite control; and continuously and dynamically adjusts ventilation through periodic rolling iterations, upgrading from traditional "controlling air temperature" to "controlling the perceived heat state of the pig population," achieving on-demand ventilation.

[0017] Wherein, in step S5, when a persistent high thermal load grid is determined, an overhead rail-mounted inspection robot is scheduled to perform intensive sampling on this area. After the persistent high thermal load area is identified, the inspection robot is scheduled to perform intensive sampling on this grid area, reduce the inspection sampling interval, acquire more thermal imaging and environmental data of this area, improve the data acquisition density of high heat risk areas, further improve the accuracy of ventilation decision-making in this area, and realize the linkage between perception and control.

[0018] In step S6, air volume-rotational speed conversion: the feedforward term performs open-loop compensation based on the fan characteristic curve and uses the motor torque to estimate the system resistance; the feedback term performs PID closed-loop correction based on the rotational speed deviation, and outputs the final rotational speed command. By adopting composite control of feedforward and feedback, the feedforward term uses the fan characteristic curve and motor torque to estimate the system resistance for open-loop compensation, so as to compensate for the rotational speed deviation caused by the change of system resistance in advance, and the feedback PID performs closed-loop correction on the deviation between the actual rotational speed and the target rotational speed. Compared with a single feedback PID, the fan rotational speed response time is significantly shortened, and the air volume can quickly track the change of thermal load.

[0019] The beneficial effects of the present invention are divided into the hardware equipment level, the perception temperature measurement level, the decision control level, the fan execution level, and the breeding application level: I. Hardware equipment level 1. Two sets of front and rear elastic floating overhead rail cleaning mechanisms are provided, the cleaning wheels are always pressed against the running surface of the overhead rail by the top springs, and the overhead rail is cleaned synchronously with the reciprocating inspection of the robot, without the need for manual high-altitude shutdown to clean the overhead rail; the cleaning wheels can adapt to the deformation of the overhead rail and local protruding sundries, and perform floating avoidance without jamming and damage; it can effectively remove pig hair, dust and feed debris on the surface of the overhead rail, avoid slipping, jamming and derailment of the traveling wheels, and significantly improve the operation reliability of the overhead rail inspection robot under the harsh high-dust working condition of the pig house; 2. The robot is equipped with front and rear anti-collision energy-absorbing boxes to buffer collision impacts and protect the cleaning mechanism and the whole machine; an internal chain transmission structure is adopted to resist dust corrosion; the storage battery supplies power on the machine to realize reciprocating inspection without trailing cables; independent heat dissipation windows are arranged for the servo motor and the storage battery to reduce the temperature rise of components in the high-temperature pig house environment and prolong the service life of the equipment; 3. The system hardware is integrated in one unit, and the overhead rail inspection robot undertakes the tasks of rail cleaning, multi-source perception and edge computing at the same time. The perception hardware is arranged on the overhead rail directly below the air supply channel, and the perception field of view covers the pig pens; the permanent magnet direct drive fan, the air supply channel and the air supply window form a zoned air supply system, forming a complete closed-loop system architecture.

[0020] II. Perception temperature measurement level 1. No longer relying on fixed-point air temperature sensors, the system achieves distributed data collection throughout the pigsty by relying on mobile inspection via overhead rails; it simultaneously collects the temperature field of pigs' bodies, the ground, walls, and equipment background temperature field; environmental sensors simultaneously collect ammonia, CO2, temperature, and humidity; visible light cameras identify pig density and posture, and optical flow methods calculate behavioral activity. 2. The multi-scale non-uniformity correction algorithm suppresses ghosting noise in thermal imaging caused by strong heat sources such as piggery heating lamps, controlling the temperature measurement error within ±0.3℃; the body surface temperature extraction based on the CBAM attention mechanism eliminates the need for manual ROI selection, adapts to various pig postures such as lying down, side-lying, and standing, and assigns differentiated weights to key physiological temperature measurement sites such as the ear base and the midline of the back; the extended Kalman filter makes optimal estimates of multi-frame temperatures, eliminating single-frame temperature fluctuations caused by airflow and obstruction; and Kriging interpolation generates a continuous air temperature interpolation field, significantly improving temperature measurement accuracy and automation.

[0021] III. Decision-making and Control Level 1. Based on the physical energy balance equation of pig heat balance, a somatosensory temperature model was constructed, and different convective heat transfer coefficient-wind speed fitting relationships were selected to distinguish between standing and lying postures; the heat stress index (TSI) was calculated using fuzzy membership degree, which is closer to the actual physiological thermal state of pigs than the traditional empirical formula. 2. Construct an m×n grid heat load distribution map, use the AHP (Analytic Hierarchy Process) to calibrate weights, and integrate heat stress index, pig density, behavioral activity, and body surface temperature shift; generate a continuous heat load surface using bicubic B-spline interpolation; maintain a cyclic buffer to store the historical heat load matrix, and calculate the rate of change of heat load to achieve feedforward early warning; 3. Multi-dimensional high heat load determination, integrating absolute threshold, relative percentile, and rate of rise, superimposed with time window duration anti-jitter constraints, and also includes special scenario determination such as pig herd aggregation, panting recognition, feeding period, extreme summer weather, and fan failure neighboring areas; identify grids with continuous high heat load, schedule inspection robots to encrypt sampling, and realize perception and control linkage; 4. Supports online parameter tuning of fuzzy PID and can also switch to MPC model predictive control; the zoned air volume allocation algorithm includes heat load ratio benchmark allocation, behavioral status air volume upper and lower limit constraints, total rebalancing, and adjacent zone smoothing constraints to suppress cross ventilation; forced basic ventilation lower limit for excessive ammonia; takes into account both the thermal comfort of the pigs and the air quality in the barn.

[0022] IV. Wind Turbine Execution Level The system uses a permanent magnet synchronous fan paired with a permanent magnet direct-drive fan driver; the airflow-speed conversion adopts a feedforward-feedback composite control; the feedforward term uses the fan characteristic curve combined with the motor torque to estimate the system resistance to achieve open-loop compensation, and the feedback PID performs closed-loop correction of speed deviation; compared with simple feedback PID, the speed response speed is greatly improved, which can quickly track the transient heat load changes in the pig house; the permanent magnet direct drive eliminates belt drive, resulting in low mechanical loss and enabling wide-range stepless speed regulation, reducing the energy consumption of the ventilation system; the air supply window is equipped with a four-way air outlet, which diffuses the airflow in four directions, improves the uniformity of the airflow field inside the pig house, reduces the local direct blowing effect, and reduces the risk of cold stress to pigs caused by cross drafts.

[0023] V. Application in Aquaculture 1. This solution shifts the focus from "controlling air temperature" to "controlling the heat-sensitive state of the pig population," effectively reducing the probability of heat and cold stress in pigs, improving pig welfare, and lowering the feed conversion ratio. 2. Achieve zoned ventilation as needed, avoid ineffective ventilation, and reduce measured ventilation energy consumption; 3. The overhead rail cleaning mechanism ensures the long-term stable operation of the inspection equipment and reduces the workload of equipment maintenance; the system supports remote monitoring, reducing the workload of feeding personnel and is suitable for large-scale closed pig houses. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a 3D structural diagram of the overhead rail cleaning mechanism; Figure 2 This is the front view of the overhead rail cleaning mechanism; Figure 3 Right view of the overhead rail cleaning mechanism; Figure 4 A three-dimensional structural diagram of the overhead rail inspection robot from a first angle; Figure 5 This is a three-dimensional structural diagram of the overhead rail inspection robot from a second angle. Figure 6 A three-dimensional structural diagram of the main body of the inspection robot from a first angle; Figure 7 This is a three-dimensional structural diagram of the main body of the inspection robot from a second angle; Figure 8 This is a diagram showing the internal structure of the main body of the inspection robot; Figure 9A three-dimensional structural diagram of a pigsty ventilation dynamic matching system for thermal imaging inspection of pig herds from the first angle. Figure 10 A three-dimensional structural diagram of the pigsty ventilation dynamic matching system for thermal imaging inspection of pig herds from a second angle. Figure 11 A three-dimensional structural diagram of the air supply window; Figure 12 A flowchart illustrating the dynamic matching process of ventilation in pig houses for thermal imaging inspections of pig herds.

[0026] Reference numerals: 1. Suspension rail; 2. Connecting horizontal seat; 3. Connecting vertical seat; 4. Vertical slide rail; 5. Vertical guide rod; 6. Slide seat; 7. Limiting block; 8. Top spring; 9. Sweeping wheel; 10. Inspection robot body; 11. Main shell; 12. Front support arm; 13. Rear support arm; 14. Drive wheel; 15. Driven wheel; 16. Servo motor; 17. Drive sprocket; 18. Driven sprocket; 19. Chain; 20. Battery; 21. Motor heat dissipation window; 22. Battery heat dissipation window; 23. Front anti-collision energy absorption box; 24. Rear anti-collision energy absorption box; 25. Sensing module; 26. Communication module; 27. Edge computing and central control module; 28. Air supply channel; 29. ​​Air supply window; 30. Air outlet. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The application scenario is a large-scale gestation sow house with a length of 100m and a width of 12m; the ceiling of the pig house is fixed with hanging rails, which are arranged directly below the air supply duct; the pig house is divided into 10 pen areas and 8 ventilation zones; the hanging rail inspection robot walks at a speed of 0.3m / s, and a single complete inspection takes 6 minutes.

[0029] Example 1, as Figures 1 to 3As shown, a suspended rail cleaning mechanism is provided. The suspended rail 1 is fixedly installed on the bottom surface of the pigsty ceiling, and the horizontal working surface under the suspended rail 1 is the running surface of the suspended rail 1. A horizontally arranged connecting cross seat 2 is provided at the bottom of the suspended rail 1; two sets of vertical connecting vertical seats 3 are fixed on the top surface of the connecting cross seat 2, and the two sets of connecting vertical seats 3 are symmetrically distributed on the left and right sides of the suspended rail 1. A vertical sliding groove 4 is opened on the connecting vertical seat 3, and a vertical guide rod 5 is fixed inside the vertical sliding groove 4; a sliding seat 6 is slidably assembled on the vertical guide rod 5; limiting blocks 7 are fixed on the left and right sides of the sliding seat 6, and the limiting blocks 7 are close to the inner wall of the connecting vertical seat 3, restricting the sliding seat 6 to slide up and down along the vertical guide rod 5 and preventing it from rotating; a top spring 8 is sleeved on the outside of the vertical guide rod 5, and the top spring 8 continuously applies a downward elastic force to the sliding seat 6; a cleaning wheel 9 is rotatably installed on the inner side of the sliding seat 6, and the wheel surface of the cleaning wheel 9 is always elastically pressed against the running surface of the suspended rail 1.

[0030] Continue as Figures 1 to 3 As shown, the working process of the overhead rail cleaning mechanism is as follows: the cleaning wheel 9 presses down on the running surface of the overhead rail 1 with the top spring 8; the mechanism moves back and forth with the robot, and the cleaning wheel 9 rolls along the running surface of the overhead rail 1 to clean and remove the pig hair, dust, and feed debris accumulated on the running surface of the overhead rail 1; when there are local protrusions or track deflection deformation in the overhead rail 1, the cleaning wheel 9 is squeezed upward, which drives the slide 6 to overcome the elastic force of the top spring 8 and slide upward to achieve floating avoidance, without jamming or damage; after the obstacle passes, the top spring 8 pushes the slide 6 to reset, and the cleaning wheel 9 presses down on the running surface again.

[0031] Example 2, as Figures 4 to 5 As shown, a suspended rail inspection robot is provided. The main body 10 of the suspended rail inspection robot is suspended on the suspended rail 1. A front anti-collision energy absorption box 23 is installed at the front end of the robot body and a rear anti-collision energy absorption box 24 is installed at the rear end. A set of suspended rail 1 cleaning mechanism is installed in front of the front anti-collision energy absorption box 23 and a set of suspended rail 1 cleaning mechanism is installed behind the rear anti-collision energy absorption box 24.

[0032] like Figures 6 to 8 As shown, the main body 10 of the inspection robot includes a main housing 11; two sets of symmetrically arranged drive wheels 14 are mounted on the front top of the main housing 11 via a front support arm 12, and the drive wheels 14 travel along the running surface of the overhead rail 1; two sets of symmetrically arranged driven wheels 15 are mounted on the rear top of the main housing 11 via a rear support arm 13. A servo motor 16 is arranged inside the main housing 11, and the output end of the servo motor 16 is connected to a drive sprocket 17; a driven sprocket 18 is arranged inside the front support arm 12, and the driven sprocket 18 is coaxial with the drive wheels 14; the drive sprocket 17 and the driven sprocket 18 are driven by a chain 19; a battery 20 is also arranged inside the main housing 11 to power all electrical components of the robot.

[0033] like Figures 4 to 5As shown, a sensing module 25 and a communication module 26 are provided at the bottom of the main housing 11. The sensing module 25 includes an infrared thermal imager, an environmental sensor group, and a visible light camera. Please refer to the prior art for the connection relationship of each part. The environmental sensor group integrates a temperature and humidity sensor, an ammonia sensor, and a CO2 sensor. The communication module 26 includes an industrial switch and a remote monitoring terminal. Please refer to the prior art for the connection relationship of each part.

[0034] Continue as Figures 4 to 5 As shown, the edge computing and central control module 27 is arranged on the top of the main housing 11, including an edge computing unit, a central controller, and a data storage unit. Please refer to the prior art for the connection relationship of each part.

[0035] like Figures 6 to 8 As shown, the bottom of the main housing 11 has heat dissipation windows, which are divided into motor heat dissipation window 21 (near the servo motor 16) and battery heat dissipation window 22 (near the battery 20), respectively to dissipate heat from the servo motor 16 and the battery 20.

[0036] like Figures 4 to 8 As shown, the working process of the overhead rail inspection robot is as follows: the battery 20 provides power, the servo motor 16 drives the active sprocket 17, the chain 19 drives the driven sprocket 18, and drives the active wheel 14 to roll along the running surface of the overhead rail 1. The driven wheel 15 follows and supports, enabling the robot to move back and forth along the overhead rail 1; when the robot moves forward, the front-end overhead rail 1 cleaning mechanism cleans the running surface of the overhead rail 1; when the robot moves backward, the rear-end overhead rail 1 cleaning mechanism cleans the running surface of the overhead rail 1; the front and rear anti-collision energy-absorbing boxes 24 buffer the impact of collisions; the bottom sensing module 25 collects images of the pigsty's internal thermal imaging, ambient gas, and pig herd; the edge computing unit completes the image and sensor data processing; the motor heat dissipation window 21 and the battery heat dissipation window 22 dissipate heat from the heat-generating components inside the shell.

[0037] Example 3, as Figures 9 to 10 As shown, a dynamic matching system for piggery ventilation based on thermal imaging inspection is provided. An air supply channel 28 is fixed on the ceiling of the piggery. A permanent magnet synchronous fan is installed at the inlet of the air supply channel 28. Multiple air supply windows 29 are arranged on the bottom surface of the air supply channel 28, penetrating the ceiling and supplying air downwards towards the piggery pens. A hanging rail 1 is fixed on the bottom surface of the ceiling, directly below the air supply channel 28. An inspection robot as described in Embodiment 2 is mounted on the hanging rail 1. The permanent magnet synchronous fan is equipped with a permanent magnet direct-drive fan driver. The permanent magnet direct-drive fan driver communicates with the central controller on the inspection robot via an industrial bus, receives control commands, drives the permanent magnet synchronous fan to adjust its speed, and transmits back the fan speed, current, and vibration status.

[0038] Example 4, as Figures 9 to 11As shown, a dynamic matching system for piggery ventilation based on thermal imaging inspection is provided. The difference between this embodiment and Embodiment 3 is that each air supply window 29 is equipped with four sets of air outlets 30, which outlet in four different directions, allowing fresh air to diffuse to multiple pens around the air supply window 29. The four-way air outlets 30 on the air supply window 29 diffuse airflow in four directions, improving the uniformity of the airflow field inside the piggery, reducing the local direct blowing effect, and lowering the risk of cold stress in pigs caused by drafts.

[0039] Example 5, as Figure 12 As shown, a dynamic matching method for piggery ventilation based on thermal imaging inspection is provided. The hardware in this embodiment adopts the system from Embodiment 4, with a thermal imager resolution of 640×512 and a thermal imaging frame rate of 10fps; the edge computing unit is an NVIDIA Jetson Xavier NX; the central controller is a Siemens S7-1200 PLC; the permanent magnet synchronous fan has a rated power of 1.5kW and a speed range of 0-600rpm; its complete workflow is as follows: S1, Multi-source data acquisition The overhead inspection robot moves back and forth along the overhead rail; the infrared thermal imager collects the thermal imaging temperature field of the pig herd, and at the same time collects the background temperature field of the ground, walls, ceiling and equipment surfaces; the environmental sensor group collects the spatial distribution of temperature and humidity, ammonia and CO2; the visible light camera collects images of the pig house, completes the counting of pigs and posture recognition, and uses the optical flow method to calculate the behavioral activity index.

[0040] S2, Thermal Imaging Multi-Algorithm Preprocessing Edge computing units perform multi-scale scene adaptive non-uniformity correction, with the following correction formula: In the formula: For the corrected grayscale; Original grayscale; It is a spatiotemporal joint correction factor; Image gradient; suppress ghost noise from heating lamps, and control temperature measurement error within ±0.3℃.

[0041] Perform adaptive extraction of pig body surface temperature based on CBAM attention mechanism: In the formula, The set of pixels segmented from the pig's body. CBAM output pixel attention weights; assign weights to different physiological parts: ear root 0.35, back midline 0.30, groin 0.20, nose mirror 0.10, tail root 0.05; no manual ROI required, adaptable to prone, side-lying, and standing postures.

[0042] Perform extended Kalman filtering and multi-frame temperature fusion: In the formula: To model based on pig metabolic rate; The process noise covariance characterizes the fluctuations in heat production in pigs; To characterize sensor noise using the observed noise covariance; output To integrate the final body surface temperature after noise reduction, a Kriging spatial interpolation algorithm is used to generate an air temperature interpolation field.

[0043] S3. Calculation of perceived temperature and heat stress index Based on the heat balance and energy balance equation for pigs: In the formula, For metabolic heat production, , The weight of the pig; Heat gain through radiation; It allows heat to evaporate from the skin; Heat dissipation through skin radiation; It allows heat to be dissipated through convection from the skin; For conduction and heat dissipation; This is for the accumulation of heat in the body.

[0044] Different convective heat transfer coefficient-wind speed fitting formulas are used to distinguish the attitude: Standing posture: ; prone position: .

[0045] Constructing equivalent perceived temperature Model; fuzzy membership functions are used to calculate the membership degrees for cold stress, comfort, and heat stress respectively, and the heat stress index is calculated: S4. Construct a heat load distribution map Divide the pigsty into The grid unit, in this embodiment, is a 10×8 grid, with each grid cell measuring 10m×1.5m; Grid-based comprehensive heat load index: In the formula: The weights are determined by the Analytic Hierarchy Process (AHP). To normalize the pig herd density; For behavioral activity level; ; .

[0046] A continuous heat load distribution surface is generated using bicubic B-spline interpolation; a cyclic buffer of length 10 is maintained to store multi-period historical heat load matrices, and the rate of change of heat load is calculated. Used for feedforward early warning.

[0047] S5. Ventilation Decisions and Zonal Airflow Allocation Multi-dimensional high heat load assessment: Simultaneously employing the absolute threshold of perceived temperature, the 90th percentile of the heat load index, and the rate of increase of heat load; setting a time window. Duration threshold Anti-shake; and simultaneously recognizes special scenarios with high heat loads, such as pig herds gathering, panting, feeding periods, extreme summer weather, and neighboring areas experiencing fan failures.

[0048] When a grid with a persistently high heat load is identified, a suspended rail inspection robot is dispatched to that area to perform encrypted sampling.

[0049] The controller can be either fuzzy PID or alternative MPC model predictive control; the fuzzy PID uses the error between the rounded average heat load index and the target heat load (0.3) and the rate of error change as inputs, and 49 fuzzy rules are tuned online. MPC employs a discrete state-space model, uses quadratic programming to solve for the optimal control sequence, and performs rolling time-domain optimization.

[0050] Zoned airflow distribution (baseline airflow): Airflow upper and lower limits are set based on the pig herd behavior (sleeping and resting, standing with slight activity, feeding, and active fighting); after constraint pruning, total volume rebalancing is performed. Apply smoothing constraints on airflow between adjacent zones to limit the airflow difference between adjacent grids and suppress cross ventilation; when the ammonia concentration is greater than 15ppm, a lower limit for the basic ventilation of the entire dormitory is forcibly set.

[0051] S6, Closed-loop ventilation implementation The air volume-speed conversion adopts a feedforward-feedback composite control; the fan model is... Feedforward speed: ,in The system resistance is estimated using the motor torque; the feedback term is corrected by PID closed-loop based on the deviation between the actual speed and the target speed, and the final speed command is sent to the permanent magnet direct drive fan driver; the permanent magnet synchronous fan performs speed regulation, and the driver returns the speed, current, and vibration status.

[0052] S7, Rolling Iteration Wait for the next inspection cycle to arrive, and repeat steps S1-S6 to continuously and dynamically match the ventilation volume of the pigsty.

[0053] Example 6 is based on Example 5. The only difference between this example and Example 5 is that the application scenario is a fattening pig house with a higher pig density. The thermal imaging sampling frame rate is increased to 20fps, and the heat dissipation coefficient in the thermal balance model is adjusted accordingly. The rest of the hardware and algorithm logic remain unchanged.

[0054] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A suspended rail cleaning mechanism, comprising a suspended rail fixed to the bottom surface of the pigsty ceiling, characterized in that, A horizontally arranged connecting cross seat is provided at the bottom of the hanging rail. Two sets of vertically arranged connecting cross seats are fixed on the top surface of the connecting cross seat, which are symmetrically distributed on the left and right sides of the hanging rail. A cleaning wheel is rotatably installed on the connecting cross seat, which is in elastic contact with the running surface of the hanging rail.

2. The overhead rail cleaning mechanism according to claim 1, characterized in that: A vertical groove is provided through the connecting vertical seat. A vertical guide rod is fixedly connected in the vertical groove. A slide block is slidably installed on the vertical guide rod. Limiting blocks are fixed on both the left and right sides of the slide block to restrict the rotation of the slide block and make it only slide up and down. A top spring is installed on the vertical guide rod to push the slide block downward. The cleaning wheel is rotatably installed on the inner side of the slide block.

3. A suspended rail inspection robot, comprising a robot body that travels along a suspended rail, characterized in that: A front anti-collision energy-absorbing box and a rear anti-collision energy-absorbing box are installed sequentially at the front and rear ends of the main body of the inspection robot. The overhead rail cleaning mechanism according to claim 2 is installed sequentially in front of the front anti-collision energy-absorbing box and behind the rear anti-collision energy-absorbing box. A sensing module and a communication module are installed at the bottom of the main body of the inspection robot, and an edge computing and central control module is installed at the top of the main body of the inspection robot.

4. The overhead rail inspection robot according to claim 3, characterized in that: The main body of the inspection robot includes a main shell. Two sets of symmetrical drive wheels that move along the overhead rail are rotatably mounted on the front top of the main shell via a front support arm. Two sets of symmetrical driven wheels that move along the overhead rail are rotatably mounted on the rear top of the main shell via a rear support arm. A servo motor and a drive sprocket driven by the servo motor are installed inside the main shell. A driven sprocket that is coaxially connected to the drive wheels is hidden inside the front support arm. The drive sprocket and the driven sprocket are connected by a chain drive. A battery that supplies power to all electrical structures is also installed inside the main shell.

5. The overhead rail inspection robot according to claim 4, characterized in that: The sensing module includes an infrared thermal imager, an environmental sensor group, and a visible light camera, wherein the environmental sensor group includes at least a temperature and humidity sensor, an ammonia sensor, and a CO2 sensor; the communication module includes an industrial switch and a remote monitoring terminal; the edge computing and central control module includes an edge computing unit, a central controller, and a data storage unit.

6. The overhead rail inspection robot according to claim 4, characterized in that: Multiple heat dissipation windows are also installed at the bottom of the main housing, including motor heat dissipation windows near the servo motor and battery heat dissipation windows near the battery.

7. A dynamic matching system for pig house ventilation based on thermal imaging inspection of pig herds, comprising an air supply duct fixed to the ceiling of the pig house, a permanent magnet synchronous fan supplying air into the duct at its inlet end, multiple air supply windows penetrating the ceiling and supplying air to the pigs located on the bottom surface of the duct, and a hanging rail fixed to the ceiling directly below the air supply duct, characterized in that: A rail inspection robot as described in claim 5 is installed on the rail, and a permanent magnet direct drive fan driver capable of receiving instructions from the rail inspection robot is connected to one side of the permanent magnet synchronous fan. Air outlets that supply air to four different directions are opened on the air supply window.

8. A method for dynamically matching ventilation in pigsties using thermal imaging inspection of pig herds, characterized in that: This method is implemented using the dynamic matching system for piggery ventilation based on thermal imaging inspection of pig herds as described in claim 7, and its working steps are as follows: S1. Multi-source data acquisition: The overhead inspection robot moves back and forth along the overhead rail, and the infrared thermal imager collects the thermal imaging temperature field of the pig herd as well as the background temperature field of the ground, walls and equipment; the environmental sensor group collects the spatial distribution of temperature, humidity, ammonia and CO2 parameters; the visible light camera collects images of the pig herd, identifies the pig herd density and posture, and calculates the behavioral activity level through optical flow method. S2. Thermal Imaging Multi-Algorithm Preprocessing: The edge computing unit performs multi-scale non-uniformity correction on the original thermal image; the pig's body surface temperature is adaptively extracted based on the CBAM attention mechanism, and differentiated temperature measurement weights are assigned to the ear root, back midline, groin, nose mirror, and tail root; then, the extended Kalman filter is used to perform optimal estimation of the body surface temperature of multiple consecutive frames, and output a stable pig body surface temperature; at the same time, Kriging interpolation is used to generate an air temperature interpolation field. S3. Calculation of perceived temperature and heat stress index: A perceived temperature model is constructed based on the physical equation of heat balance of pigs. The corresponding convective heat transfer coefficient-wind speed fitting relationship is selected according to the standing or lying posture of pigs. The membership degree of cold, comfort and heat stress is calculated by using fuzzy membership function, and the heat stress index TSI is solved. S4. Construct a heat load distribution map: Divide the pigsty space into m×n grid units; Pig density and behavioral activity were statistically analyzed grid by grid, and the comprehensive heat load index of the grid was calculated; a heat load distribution surface was generated by bicubic B-spline interpolation. Maintain a circular buffer to store the historical heat load matrix for multiple periods, and calculate the rate of change of heat load as a basis for feedforward early warning; S5. Ventilation Decision and Zoned Airflow Allocation: Performs multi-dimensional high heat load judgment, combining absolute threshold, relative percentile, heat load rise rate, superimposed time window duration anti-jitter constraint, and identifies high heat load in special scenarios. The system calls a fuzzy PID controller or an alternative MPC model predictive controller to calculate the total target air volume for the entire shed; it then performs zoned air volume allocation: the baseline air volume is obtained based on the heat load ratio, and upper and lower limits of air volume are applied in combination with the pig herd behavior status to complete the total rebalancing, and air volume smoothing constraints are applied to adjacent zones; when the ammonia concentration exceeds the threshold, the basic ventilation lower limit is forcibly set. S6. Closed-loop ventilation execution: The target air volume of each zone is converted into a permanent magnet synchronous fan speed command through feedforward-feedback composite control and sent to the driver; the permanent magnet synchronous fan performs speed regulation and transmits back the speed, current and vibration status. S7. Rolling Iteration: Enter the next round of inspection cycle, repeat steps S1-S6, and continuously and dynamically match the ventilation volume of the pigsty.

9. The method for dynamically matching piggery ventilation during thermal imaging inspection of pig herds according to claim 8, characterized in that: In step S5, when a grid with a persistent high heat load is determined, a rail-mounted inspection robot is dispatched to perform intensive sampling of the area.

10. The method for dynamically matching piggery ventilation during thermal imaging inspection of pig herds according to claim 8, characterized in that: Step S6 Airflow-Speed ​​Conversion: The feedforward term uses the fan characteristic curve and the motor torque to estimate the system resistance for open-loop compensation; the feedback term uses the speed deviation PID closed-loop correction to output the final speed command.

Images